NexisDocs

Product changelog / May–July 2026

See what’s new in Nexis.

New ways to build apps, run agents, research the web, manage wallets, and measure autonomous systems—organized by release month.

Current release

July 2026

Read the latest

Latest

July 2026

Build complete apps, move between active agent sessions, research with stronger fallbacks, and manage user-owned wallets from one clearer workspace.

NEXBENCH tasks
214
benchmark categories
8
trials per task
5×

01

Benchmarking

NEXBENCH 2.1 is now public

Builders can measure whether an autonomous Web3 agent is both capable and safe across execution, swaps, bridging, DeFi, research, security, portfolio analysis, and governance.

  • A public leaderboard, agent comparison view, submission flow, and methodology guide now share the same hash-sealed run manifests.
  • The zero-runtime-dependency CLI can scaffold an adapter, run the public suite, verify results, mint a valid manifest, and submit it for review.
  • Deterministic forks, frozen research corpora, safety-zeroed scoring, and programmatic graders keep results reproducible and harder to game.

02

Building

Build complete web and mobile apps

Describe what you want, follow the build in real time, then inspect the generated files and working preview in one workspace.

  • Prompts can produce responsive web apps and Expo projects, with separate iOS and Android export records for mobile builds.
  • Source files, run progress, verification results, and the latest preview stay attached to the project after reload.
  • Landing-page requests now generate purpose-built marketing sites instead of a generic starter app.

03

Agent workspace

Move between active chats without losing your place

Current agent sessions stay within reach while each conversation keeps its own context, progress, and approval state.

  • The header shows recent sessions with clear states for work in progress, completion, failure, and approvals that need attention.
  • Reopening a chat preserves its session so follow-up messages continue in the right conversation.
  • The composer, sidebar, account controls, and activity indicators were refined to make longer-running work easier to follow.

04

Research

Research that adapts to the request

Nexis can now choose between fast search, extracted sources, deeper corroboration, direct page reading, and multi-page research.

  • Search supports news, images, places, shopping, scholarly sources, patents, and other specialized result types.
  • Results can follow your chosen region and language for more relevant research.
  • If one search provider is unavailable, Nexis switches routes automatically and preserves source links for review.

05

Commerce

Agent products and x402 payments

Nexis added the foundation for agents to discover products, coordinate paid capabilities, and retain payment state as part of a run.

  • The Agent Studio now exposes product, template, pack, learning, analytics, and control surfaces from the main app navigation.
  • x402 payment records preserve status history so a paid agent interaction can be audited after completion.
  • Marketplace pages reuse the landing system, keeping discovery, execution, and product education visually connected.

06

Developer experience

Documentation and research became first-class

A complete developer portal now covers local setup, platform concepts, tutorials, operations, and NEXBENCH from one searchable shell.

  • Guides cover streamed runs, custom MCP tools, sandbox artifacts, on-chain intents, deployment, retries, rate limits, and troubleshooting.
  • The research and benchmark surfaces now use the same chapter-based visual language as the home page.
  • The changelog joins the docs navigation so product progress stays discoverable beside the implementation reference.

07

Wallet safety

Secure export for user-owned wallets

Eligible user-owned Turnkey wallets can now start a secure export from the account menu, while agent and delegated wallets remain non-exportable.

  • Nexis explains when export is unavailable instead of presenting a control that cannot complete.
  • Step-up authentication and Turnkey’s secure export flow keep wallet material out of chat and away from agents.
  • Optional spend limits remain explicit instead of silently becoming unlimited permissions.

Platform release

June 2026

The June release focused on making the core agent experience faster, more durable, and easier to trust across chat, automation, wallets, and production use.

agent command world
3D
event-driven automation
24/7
durable run ledger
1

01

Intelligence

Persistent memory and sharper tool routing

Conversations can now retain stable session context while the runtime chooses tools and models from the actual intent of the request.

  • Lazy compaction and semantic recall keep long-running conversations useful without sending the entire history on every turn.
  • Intent-precision routing, parallel tool use, fallbacks, and self-healing reduce empty answers and avoid unnecessary provider calls.
  • Tool-selection activity and cited sources are visible while the response streams.

02

Automation

Event-driven agent workflows

Agents can react to governed external events instead of waiting for a user to reopen the app and type a new command.

  • Signed webhook ingestion, scheduled triggers, event inboxes, notification outboxes, and circuit breakers form one reliable workflow.
  • Market and exchange updates can create a run with a typed input template and trace the resulting notification back to the source event.
  • Every triggered run remains subject to authorization, policy, approval, and audit controls.

03

Experience

Agent World and a calmer chat surface

The app gained a spatial side panel where users can see agents work, alongside a cleaner transcript and more compact navigation.

  • Player avatars, movement, chat, activity states, and wardrobe progression make concurrent agent work easier to scan.
  • Inline price-chart artifacts, full-screen artifact viewing, and Perplexity-style response typography keep evidence close to the answer.
  • The sidebar can recede after a conversation starts while preserving quick access to agents and history.

04

On-chain

Safer contract and exchange workflows

On-chain operations gained deeper verification and more complete reconciliation across contracts, Solana programs, and exchange activity.

  • Proxy implementation checks and verification packets make smart-contract review results durable and inspectable.
  • Solana deployment is represented as a planned, approval-gated intent rather than an unrestricted signing action.
  • Exchange and Yellowstone streams can reconcile activity back into portfolio and automation state.

05

Reliability

Production fallbacks without hidden failure

The web app and runtime became more resilient to transient reads, provider outages, database pressure, and interrupted streams.

  • Portfolio, agent, wallet, and optional app reads degrade deliberately while preserving actionable error state.
  • Agent creation and publishing use isolated write paths and atomic version state to avoid partial saves.
  • Production smoke routes now cover authenticated chat, wallets, runs, and database readiness.

Public foundation

May 2026

The first public platform foundation brought agents, wallets, markets, connectors, artifacts, voice, and approvals into one responsive product.

responsive control surface
Web
realtime agent mode
Voice
connector gateway
MCP

01

Core platform

One place to create and run agents

Nexis launched with authenticated onboarding, agent creation, streamed runs, specialist coordination, and a durable event timeline.

  • Users can create agents, choose models and tools, start a run, and watch status, tool calls, checkpoints, and sources arrive in real time.
  • The ManagerAgent coordinates bounded research, portfolio, risk, compliance, and coding specialists while remaining responsible for the answer.
  • Responsive app, mobile, and settings surfaces keep the same workflows usable from smaller screens.

02

Wallets & markets

Embedded wallets with live portfolio context

Users can provision embedded wallets, inspect balances and market data, and prepare sends, swaps, bridges, and liquidity actions from chat.

  • Portfolio views include token grouping, price history, watchlists, asset detail pages, and live market data.
  • Confirmed sends, swaps, and bridges reconcile receipts back into activity and balance-refresh state.
  • EVM and supported non-EVM accounts share typed intent and approval boundaries instead of exposing raw transaction signing.

03

Connectors

External tools without exposing credentials

A connector catalog and MCP gateway let agents reach approved services while provider tokens remain server-side.

  • Users can discover and connect Pipedream integrations, iMessage, Coinbase, Robinhood, Blockscout, Firecrawl, and other scoped tools.
  • Composer controls show available connectors and make their activation state clear before a run starts.
  • Connector calls pass through scope checks, policy checks, timeouts, audit events, and approval rules.

04

Artifacts

Runs can deliver real work products

Agent output moved beyond chat text into versioned artifacts that can be previewed, resumed, and traced back to the run that created them.

  • Code, PDF, image, research, and sandbox outputs have first-class records and scoped access.
  • Generated work is scanned and published through the artifact service instead of leaking directly from a sandbox.
  • Approval resume and persisted context let interrupted work continue without losing its output trail.

05

Voice & messaging

Nexis moved beyond the browser composer

Realtime voice and iMessage introduced faster ways to ask for portfolio context, research, and governed wallet workflows.

  • LiveKit-based voice sessions support microphone input, audio playback, portfolio context, and approved browser-command handoff.
  • iMessage connection flows gained QR setup, persistent session context, typing state, and reliable inbound and outbound recovery.
  • Voice and messaging actions still surface approvals before a consequential command can continue.

06

Plans & controls

Usage, billing, and approvals became visible

Subscription entitlements, model costs, connector limits, sandbox usage, and overage controls are reflected in the product instead of being hidden infrastructure.

  • Users can inspect credit costs, manage spend limits, and choose whether overages are allowed.
  • Premium models, connector tiers, artifact storage, and sandbox resume behavior fail closed when an entitlement is missing.
  • Approval resolution is idempotent and can resume a paused run without duplicating the underlying action.

Build with Nexis

Explore the platform in the docs, or run NEXBENCH against your own agent.

Developer docsNEXBENCH on GitHub